TuneKit

Fine-tune a small language model on your own dataset and test it — all from a simple browser UI. No code required.

Main page

---

Supported Models

Model	Parameters	HuggingFace ID
DeepSeek-R1	1.5B	deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
Qwen2.5	1.5B	Qwen/Qwen2.5-1.5B
SmolLM2	1.7B	HuggingFaceTB/SmolLM2-1.7B

All models download automatically from Hugging Face on first use.

Fine-tuning Techniques

Technique	Description
LoRA	Trains a small set of adapter weights (~1% of params). Fast, low memory.
QLoRA	LoRA on a 4-bit quantized model. Lowest memory — requires a CUDA GPU.
Prefix Tuning	Prepends trainable tokens to the input. No model weights modified.
Full Fine-tuning	Updates all model parameters. Highest quality, highest memory.

---

Project Structure

tunekit/
├── pyproject.toml
├── data.json                   ← sample training data
└── src/tunekit/
    ├── trainer.py              ← fine-tuning logic (LoRA, QLoRA, Prefix Tuning, Full)
    ├── bot.py                  ← inference / chat logic
    └── app.py                  ← Streamlit UI

---

Requirements

• Python 3.10+
• uv package manager
• CUDA GPU — required only for QLoRA (MPS/CPU supported for all other techniques)

---

Setup

1. Install uv (if not already installed):

# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# or via Homebrew
brew install uv

2. Clone and enter the project directory:

cd /path/to/tunekit

3. Create a virtual environment and install dependencies:

uv venv
source .venv/bin/activate       # macOS / Linux
# .venv\Scripts\activate        # Windows

uv pip install -e .

---

Running the App

tunekit

Then open your browser at http://localhost:8501.

---

Using the UI

Left panel — Fine-tune

1. Select a model — choose from DeepSeek-R1, Qwen2.5, or SmolLM2.
2. Select a technique — LoRA, QLoRA, Prefix Tuning, or Full Fine-tuning.
3. Choose hyperparameters — pick a preset (Default / Fast / High Quality).
4. Load training data — use the bundled data.json or upload your own JSON file.
5. Click Fine-tune — training runs in the background with a live progress bar.

When training finishes, the model is saved to ./finetuned-model/ and a .zip download is offered.

Right panel — Test Bot

1. Upload the .zip produced after training, then click Load Model.
2. Type a prompt in the chat input and press Enter.
3. The bot responds using the fine-tuned model.

---

Training Data Format

data.json is a list of instruction/response pairs:

[
  {
    "instruction": "What is the capital of France?",
    "response": "The capital of France is Paris."
  },
  {
    "instruction": "Explain machine learning in one sentence.",
    "response": "Machine learning is a method of teaching computers to learn patterns from data."
  }
]

A minimum of 2 examples is required. You can upload your own file via the UI.